1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_ipfw.h" /* for ipfw_fwd */ 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 40 #include <sys/param.h> 41 #include <sys/kernel.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/proc.h> /* for proc0 declaration */ 45 #include <sys/protosw.h> 46 #include <sys/signalvar.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/sysctl.h> 50 #include <sys/syslog.h> 51 #include <sys/systm.h> 52 53 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 54 55 #include <vm/uma.h> 56 57 #include <net/if.h> 58 #include <net/route.h> 59 60 #include <netinet/in.h> 61 #include <netinet/in_pcb.h> 62 #include <netinet/in_systm.h> 63 #include <netinet/in_var.h> 64 #include <netinet/ip.h> 65 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 66 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 67 #include <netinet/ip_var.h> 68 #include <netinet/ip_options.h> 69 #include <netinet/ip6.h> 70 #include <netinet/icmp6.h> 71 #include <netinet6/in6_pcb.h> 72 #include <netinet6/ip6_var.h> 73 #include <netinet6/nd6.h> 74 #include <netinet/tcp.h> 75 #include <netinet/tcp_fsm.h> 76 #include <netinet/tcp_seq.h> 77 #include <netinet/tcp_timer.h> 78 #include <netinet/tcp_var.h> 79 #include <netinet6/tcp6_var.h> 80 #include <netinet/tcpip.h> 81 #ifdef TCPDEBUG 82 #include <netinet/tcp_debug.h> 83 #endif /* TCPDEBUG */ 84 85 #ifdef IPSEC 86 #include <netipsec/ipsec.h> 87 #include <netipsec/ipsec6.h> 88 #endif /*IPSEC*/ 89 90 #include <machine/in_cksum.h> 91 92 #include <security/mac/mac_framework.h> 93 94 static const int tcprexmtthresh = 3; 95 96 struct tcpstat tcpstat; 97 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 98 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 99 100 static int tcp_log_in_vain = 0; 101 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 102 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); 103 104 static int blackhole = 0; 105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 106 &blackhole, 0, "Do not send RST on segments to closed ports"); 107 108 int tcp_delack_enabled = 1; 109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 110 &tcp_delack_enabled, 0, 111 "Delay ACK to try and piggyback it onto a data packet"); 112 113 static int drop_synfin = 0; 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 115 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 116 117 static int tcp_do_rfc3042 = 1; 118 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 119 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 120 121 static int tcp_do_rfc3390 = 1; 122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 123 &tcp_do_rfc3390, 0, 124 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 125 126 static int tcp_insecure_rst = 0; 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 128 &tcp_insecure_rst, 0, 129 "Follow the old (insecure) criteria for accepting RST packets"); 130 131 int tcp_do_autorcvbuf = 1; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 133 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 134 135 int tcp_autorcvbuf_inc = 16*1024; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 137 &tcp_autorcvbuf_inc, 0, 138 "Incrementor step size of automatic receive buffer"); 139 140 int tcp_autorcvbuf_max = 256*1024; 141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 142 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 143 144 struct inpcbhead tcb; 145 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 146 struct inpcbinfo tcbinfo; 147 148 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 149 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 150 struct socket *, struct tcpcb *, int, int); 151 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 152 struct tcpcb *, int, int); 153 static void tcp_pulloutofband(struct socket *, 154 struct tcphdr *, struct mbuf *, int); 155 static void tcp_xmit_timer(struct tcpcb *, int); 156 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 157 158 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 159 #ifdef INET6 160 #define ND6_HINT(tp) \ 161 do { \ 162 if ((tp) && (tp)->t_inpcb && \ 163 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 164 nd6_nud_hint(NULL, NULL, 0); \ 165 } while (0) 166 #else 167 #define ND6_HINT(tp) 168 #endif 169 170 /* 171 * Indicate whether this ack should be delayed. We can delay the ack if 172 * - there is no delayed ack timer in progress and 173 * - our last ack wasn't a 0-sized window. We never want to delay 174 * the ack that opens up a 0-sized window and 175 * - delayed acks are enabled or 176 * - this is a half-synchronized T/TCP connection. 177 */ 178 #define DELAY_ACK(tp) \ 179 ((!tcp_timer_active(tp, TT_DELACK) && \ 180 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 181 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 182 183 184 /* 185 * TCP input handling is split into multiple parts: 186 * tcp6_input is a thin wrapper around tcp_input for the extended 187 * ip6_protox[] call format in ip6_input 188 * tcp_input handles primary segment validation, inpcb lookup and 189 * SYN processing on listen sockets 190 * tcp_do_segment processes the ACK and text of the segment for 191 * establishing, established and closing connections 192 */ 193 #ifdef INET6 194 int 195 tcp6_input(struct mbuf **mp, int *offp, int proto) 196 { 197 struct mbuf *m = *mp; 198 struct in6_ifaddr *ia6; 199 200 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 201 202 /* 203 * draft-itojun-ipv6-tcp-to-anycast 204 * better place to put this in? 205 */ 206 ia6 = ip6_getdstifaddr(m); 207 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 208 struct ip6_hdr *ip6; 209 210 ip6 = mtod(m, struct ip6_hdr *); 211 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 212 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 213 return IPPROTO_DONE; 214 } 215 216 tcp_input(m, *offp); 217 return IPPROTO_DONE; 218 } 219 #endif 220 221 void 222 tcp_input(struct mbuf *m, int off0) 223 { 224 struct tcphdr *th; 225 struct ip *ip = NULL; 226 struct ipovly *ipov; 227 struct inpcb *inp = NULL; 228 struct tcpcb *tp = NULL; 229 struct socket *so = NULL; 230 u_char *optp = NULL; 231 int optlen = 0; 232 int len, tlen, off; 233 int drop_hdrlen; 234 int thflags; 235 int rstreason = 0; /* For badport_bandlim accounting purposes */ 236 #ifdef IPFIREWALL_FORWARD 237 struct m_tag *fwd_tag; 238 #endif 239 #ifdef INET6 240 struct ip6_hdr *ip6 = NULL; 241 int isipv6; 242 #else 243 const void *ip6 = NULL; 244 const int isipv6 = 0; 245 #endif 246 struct tcpopt to; /* options in this segment */ 247 char *s = NULL; /* address and port logging */ 248 249 #ifdef TCPDEBUG 250 /* 251 * The size of tcp_saveipgen must be the size of the max ip header, 252 * now IPv6. 253 */ 254 u_char tcp_saveipgen[IP6_HDR_LEN]; 255 struct tcphdr tcp_savetcp; 256 short ostate = 0; 257 #endif 258 259 #ifdef INET6 260 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 261 #endif 262 263 to.to_flags = 0; 264 tcpstat.tcps_rcvtotal++; 265 266 if (isipv6) { 267 #ifdef INET6 268 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 269 ip6 = mtod(m, struct ip6_hdr *); 270 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 271 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 272 tcpstat.tcps_rcvbadsum++; 273 goto drop; 274 } 275 th = (struct tcphdr *)((caddr_t)ip6 + off0); 276 277 /* 278 * Be proactive about unspecified IPv6 address in source. 279 * As we use all-zero to indicate unbounded/unconnected pcb, 280 * unspecified IPv6 address can be used to confuse us. 281 * 282 * Note that packets with unspecified IPv6 destination is 283 * already dropped in ip6_input. 284 */ 285 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 286 /* XXX stat */ 287 goto drop; 288 } 289 #else 290 th = NULL; /* XXX: Avoid compiler warning. */ 291 #endif 292 } else { 293 /* 294 * Get IP and TCP header together in first mbuf. 295 * Note: IP leaves IP header in first mbuf. 296 */ 297 if (off0 > sizeof (struct ip)) { 298 ip_stripoptions(m, (struct mbuf *)0); 299 off0 = sizeof(struct ip); 300 } 301 if (m->m_len < sizeof (struct tcpiphdr)) { 302 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 303 == NULL) { 304 tcpstat.tcps_rcvshort++; 305 return; 306 } 307 } 308 ip = mtod(m, struct ip *); 309 ipov = (struct ipovly *)ip; 310 th = (struct tcphdr *)((caddr_t)ip + off0); 311 tlen = ip->ip_len; 312 313 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 314 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 315 th->th_sum = m->m_pkthdr.csum_data; 316 else 317 th->th_sum = in_pseudo(ip->ip_src.s_addr, 318 ip->ip_dst.s_addr, 319 htonl(m->m_pkthdr.csum_data + 320 ip->ip_len + 321 IPPROTO_TCP)); 322 th->th_sum ^= 0xffff; 323 #ifdef TCPDEBUG 324 ipov->ih_len = (u_short)tlen; 325 ipov->ih_len = htons(ipov->ih_len); 326 #endif 327 } else { 328 /* 329 * Checksum extended TCP header and data. 330 */ 331 len = sizeof (struct ip) + tlen; 332 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 333 ipov->ih_len = (u_short)tlen; 334 ipov->ih_len = htons(ipov->ih_len); 335 th->th_sum = in_cksum(m, len); 336 } 337 if (th->th_sum) { 338 tcpstat.tcps_rcvbadsum++; 339 goto drop; 340 } 341 /* Re-initialization for later version check */ 342 ip->ip_v = IPVERSION; 343 } 344 345 /* 346 * Check that TCP offset makes sense, 347 * pull out TCP options and adjust length. XXX 348 */ 349 off = th->th_off << 2; 350 if (off < sizeof (struct tcphdr) || off > tlen) { 351 tcpstat.tcps_rcvbadoff++; 352 goto drop; 353 } 354 tlen -= off; /* tlen is used instead of ti->ti_len */ 355 if (off > sizeof (struct tcphdr)) { 356 if (isipv6) { 357 #ifdef INET6 358 IP6_EXTHDR_CHECK(m, off0, off, ); 359 ip6 = mtod(m, struct ip6_hdr *); 360 th = (struct tcphdr *)((caddr_t)ip6 + off0); 361 #endif 362 } else { 363 if (m->m_len < sizeof(struct ip) + off) { 364 if ((m = m_pullup(m, sizeof (struct ip) + off)) 365 == NULL) { 366 tcpstat.tcps_rcvshort++; 367 return; 368 } 369 ip = mtod(m, struct ip *); 370 ipov = (struct ipovly *)ip; 371 th = (struct tcphdr *)((caddr_t)ip + off0); 372 } 373 } 374 optlen = off - sizeof (struct tcphdr); 375 optp = (u_char *)(th + 1); 376 } 377 thflags = th->th_flags; 378 379 /* 380 * Convert TCP protocol specific fields to host format. 381 */ 382 th->th_seq = ntohl(th->th_seq); 383 th->th_ack = ntohl(th->th_ack); 384 th->th_win = ntohs(th->th_win); 385 th->th_urp = ntohs(th->th_urp); 386 387 /* 388 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 389 */ 390 drop_hdrlen = off0 + off; 391 392 /* 393 * Locate pcb for segment. 394 */ 395 INP_INFO_WLOCK(&tcbinfo); 396 findpcb: 397 INP_INFO_WLOCK_ASSERT(&tcbinfo); 398 #ifdef IPFIREWALL_FORWARD 399 /* 400 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 401 */ 402 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 403 404 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 405 struct sockaddr_in *next_hop; 406 407 next_hop = (struct sockaddr_in *)(fwd_tag+1); 408 /* 409 * Transparently forwarded. Pretend to be the destination. 410 * already got one like this? 411 */ 412 inp = in_pcblookup_hash(&tcbinfo, 413 ip->ip_src, th->th_sport, 414 ip->ip_dst, th->th_dport, 415 0, m->m_pkthdr.rcvif); 416 if (!inp) { 417 /* It's new. Try to find the ambushing socket. */ 418 inp = in_pcblookup_hash(&tcbinfo, 419 ip->ip_src, th->th_sport, 420 next_hop->sin_addr, 421 next_hop->sin_port ? 422 ntohs(next_hop->sin_port) : 423 th->th_dport, 424 INPLOOKUP_WILDCARD, 425 m->m_pkthdr.rcvif); 426 } 427 /* Remove the tag from the packet. We don't need it anymore. */ 428 m_tag_delete(m, fwd_tag); 429 } else 430 #endif /* IPFIREWALL_FORWARD */ 431 { 432 if (isipv6) { 433 #ifdef INET6 434 inp = in6_pcblookup_hash(&tcbinfo, 435 &ip6->ip6_src, th->th_sport, 436 &ip6->ip6_dst, th->th_dport, 437 INPLOOKUP_WILDCARD, 438 m->m_pkthdr.rcvif); 439 #endif 440 } else 441 inp = in_pcblookup_hash(&tcbinfo, 442 ip->ip_src, th->th_sport, 443 ip->ip_dst, th->th_dport, 444 INPLOOKUP_WILDCARD, 445 m->m_pkthdr.rcvif); 446 } 447 448 #ifdef IPSEC 449 #ifdef INET6 450 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) { 451 ipsec6stat.in_polvio++; 452 goto dropunlock; 453 } else 454 #endif /* INET6 */ 455 if (inp != NULL && ipsec4_in_reject(m, inp)) { 456 ipsec4stat.in_polvio++; 457 goto dropunlock; 458 } 459 #endif /* IPSEC */ 460 461 /* 462 * If the INPCB does not exist then all data in the incoming 463 * segment is discarded and an appropriate RST is sent back. 464 */ 465 if (inp == NULL) { 466 /* 467 * Log communication attempts to ports that are not 468 * in use. 469 */ 470 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 471 tcp_log_in_vain == 2) { 472 if ((s = tcp_log_addrs(NULL, th, (void *)ip, ip6))) 473 log(LOG_INFO, "%s; %s: Connection attempt " 474 "to closed port\n", s, __func__); 475 } 476 /* 477 * When blackholing do not respond with a RST but 478 * completely ignore the segment and drop it. 479 */ 480 if ((blackhole == 1 && (thflags & TH_SYN)) || 481 blackhole == 2) 482 goto dropunlock; 483 484 rstreason = BANDLIM_RST_CLOSEDPORT; 485 goto dropwithreset; 486 } 487 INP_LOCK(inp); 488 489 /* 490 * Check the minimum TTL for socket. 491 */ 492 if (inp->inp_ip_minttl != 0) { 493 #ifdef INET6 494 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 495 goto dropunlock; 496 else 497 #endif 498 if (inp->inp_ip_minttl > ip->ip_ttl) 499 goto dropunlock; 500 } 501 502 /* 503 * A previous connection in TIMEWAIT state is supposed to catch 504 * stray or duplicate segments arriving late. If this segment 505 * was a legitimate new connection attempt the old INPCB gets 506 * removed and we can try again to find a listening socket. 507 */ 508 if (inp->inp_vflag & INP_TIMEWAIT) { 509 if (thflags & TH_SYN) 510 tcp_dooptions(&to, optp, optlen, TO_SYN); 511 /* 512 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 513 */ 514 if (tcp_twcheck(inp, &to, th, m, tlen)) 515 goto findpcb; 516 INP_INFO_WUNLOCK(&tcbinfo); 517 return; 518 } 519 /* 520 * The TCPCB may no longer exist if the connection is winding 521 * down or it is in the CLOSED state. Either way we drop the 522 * segment and send an appropriate response. 523 */ 524 tp = intotcpcb(inp); 525 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 526 rstreason = BANDLIM_RST_CLOSEDPORT; 527 goto dropwithreset; 528 } 529 530 #ifdef MAC 531 INP_LOCK_ASSERT(inp); 532 if (mac_check_inpcb_deliver(inp, m)) 533 goto dropunlock; 534 #endif 535 so = inp->inp_socket; 536 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 537 #ifdef TCPDEBUG 538 if (so->so_options & SO_DEBUG) { 539 ostate = tp->t_state; 540 if (isipv6) { 541 #ifdef INET6 542 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 543 #endif 544 } else 545 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 546 tcp_savetcp = *th; 547 } 548 #endif 549 /* 550 * When the socket is accepting connections (the INPCB is in LISTEN 551 * state) we look into the SYN cache if this is a new connection 552 * attempt or the completion of a previous one. 553 */ 554 if (so->so_options & SO_ACCEPTCONN) { 555 struct in_conninfo inc; 556 557 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 558 "tp not listening", __func__)); 559 560 bzero(&inc, sizeof(inc)); 561 inc.inc_isipv6 = isipv6; 562 #ifdef INET6 563 if (isipv6) { 564 inc.inc6_faddr = ip6->ip6_src; 565 inc.inc6_laddr = ip6->ip6_dst; 566 } else 567 #endif 568 { 569 inc.inc_faddr = ip->ip_src; 570 inc.inc_laddr = ip->ip_dst; 571 } 572 inc.inc_fport = th->th_sport; 573 inc.inc_lport = th->th_dport; 574 575 /* 576 * Check for an existing connection attempt in syncache if 577 * the flag is only ACK. A successful lookup creates a new 578 * socket appended to the listen queue in SYN_RECEIVED state. 579 */ 580 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 581 /* 582 * Parse the TCP options here because 583 * syncookies need access to the reflected 584 * timestamp. 585 */ 586 tcp_dooptions(&to, optp, optlen, 0); 587 /* 588 * NB: syncache_expand() doesn't unlock 589 * inp and tcpinfo locks. 590 */ 591 if (!syncache_expand(&inc, &to, th, &so, m)) { 592 /* 593 * No syncache entry or ACK was not 594 * for our SYN/ACK. Send a RST. 595 * NB: syncache did its own logging 596 * of the failure cause. 597 */ 598 rstreason = BANDLIM_RST_OPENPORT; 599 goto dropwithreset; 600 } 601 if (so == NULL) { 602 /* 603 * We completed the 3-way handshake 604 * but could not allocate a socket 605 * either due to memory shortage, 606 * listen queue length limits or 607 * global socket limits. Send RST 608 * or wait and have the remote end 609 * retransmit the ACK for another 610 * try. 611 */ 612 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 613 log(LOG_DEBUG, "%s; %s: Listen socket: " 614 "Socket allocation failed due to " 615 "limits or memory shortage, %s\n", 616 s, __func__, (tcp_sc_rst_sock_fail ? 617 "sending RST" : "try again")); 618 if (tcp_sc_rst_sock_fail) { 619 rstreason = BANDLIM_UNLIMITED; 620 goto dropwithreset; 621 } else 622 goto dropunlock; 623 } 624 /* 625 * Socket is created in state SYN_RECEIVED. 626 * Unlock the listen socket, lock the newly 627 * created socket and update the tp variable. 628 */ 629 INP_UNLOCK(inp); /* listen socket */ 630 inp = sotoinpcb(so); 631 INP_LOCK(inp); /* new connection */ 632 tp = intotcpcb(inp); 633 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 634 ("%s: ", __func__)); 635 /* 636 * Process the segment and the data it 637 * contains. tcp_do_segment() consumes 638 * the mbuf chain and unlocks the inpcb. 639 */ 640 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen); 641 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 642 return; 643 } 644 /* 645 * Segment flag validation for new connection attempts: 646 * 647 * Our (SYN|ACK) response was rejected. 648 * Check with syncache and remove entry to prevent 649 * retransmits. 650 */ 651 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 652 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 653 log(LOG_DEBUG, "%s; %s: Listen socket: " 654 "Our SYN|ACK was rejected, connection " 655 "attempt aborted by remote endpoint\n", 656 s, __func__); 657 syncache_chkrst(&inc, th); 658 goto dropunlock; 659 } 660 /* 661 * Spurious RST. Ignore. 662 */ 663 if (thflags & TH_RST) { 664 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 665 log(LOG_DEBUG, "%s; %s: Listen socket: " 666 "Spurious RST, segment rejected\n", 667 s, __func__); 668 goto dropunlock; 669 } 670 /* 671 * We can't do anything without SYN. 672 */ 673 if ((thflags & TH_SYN) == 0) { 674 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 675 log(LOG_DEBUG, "%s; %s: Listen socket: " 676 "SYN is missing, segment rejected\n", 677 s, __func__); 678 tcpstat.tcps_badsyn++; 679 goto dropunlock; 680 } 681 /* 682 * (SYN|ACK) is bogus on a listen socket. 683 */ 684 if (thflags & TH_ACK) { 685 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 686 log(LOG_DEBUG, "%s; %s: Listen socket: " 687 "SYN|ACK invalid, segment rejected\n", 688 s, __func__); 689 syncache_badack(&inc); /* XXX: Not needed! */ 690 tcpstat.tcps_badsyn++; 691 rstreason = BANDLIM_RST_OPENPORT; 692 goto dropwithreset; 693 } 694 /* 695 * If the drop_synfin option is enabled, drop all 696 * segments with both the SYN and FIN bits set. 697 * This prevents e.g. nmap from identifying the 698 * TCP/IP stack. 699 * XXX: Poor reasoning. nmap has other methods 700 * and is constantly refining its stack detection 701 * strategies. 702 * XXX: This is a violation of the TCP specification 703 * and was used by RFC1644. 704 */ 705 if ((thflags & TH_FIN) && drop_synfin) { 706 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 707 log(LOG_DEBUG, "%s; %s: Listen socket: " 708 "SYN|FIN segment rejected (based on " 709 "sysctl setting)\n", s, __func__); 710 tcpstat.tcps_badsyn++; 711 goto dropunlock; 712 } 713 /* 714 * Segment's flags are (SYN) or (SYN|FIN). 715 * 716 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 717 * as they do not affect the state of the TCP FSM. 718 * The data pointed to by TH_URG and th_urp is ignored. 719 */ 720 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 721 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 722 KASSERT(thflags & (TH_SYN), 723 ("%s: Listen socket: TH_SYN not set", __func__)); 724 #ifdef INET6 725 /* 726 * If deprecated address is forbidden, 727 * we do not accept SYN to deprecated interface 728 * address to prevent any new inbound connection from 729 * getting established. 730 * When we do not accept SYN, we send a TCP RST, 731 * with deprecated source address (instead of dropping 732 * it). We compromise it as it is much better for peer 733 * to send a RST, and RST will be the final packet 734 * for the exchange. 735 * 736 * If we do not forbid deprecated addresses, we accept 737 * the SYN packet. RFC2462 does not suggest dropping 738 * SYN in this case. 739 * If we decipher RFC2462 5.5.4, it says like this: 740 * 1. use of deprecated addr with existing 741 * communication is okay - "SHOULD continue to be 742 * used" 743 * 2. use of it with new communication: 744 * (2a) "SHOULD NOT be used if alternate address 745 * with sufficient scope is available" 746 * (2b) nothing mentioned otherwise. 747 * Here we fall into (2b) case as we have no choice in 748 * our source address selection - we must obey the peer. 749 * 750 * The wording in RFC2462 is confusing, and there are 751 * multiple description text for deprecated address 752 * handling - worse, they are not exactly the same. 753 * I believe 5.5.4 is the best one, so we follow 5.5.4. 754 */ 755 if (isipv6 && !ip6_use_deprecated) { 756 struct in6_ifaddr *ia6; 757 758 if ((ia6 = ip6_getdstifaddr(m)) && 759 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 760 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 761 log(LOG_DEBUG, "%s; %s: Listen socket: " 762 "Connection attempt to deprecated " 763 "IPv6 address rejected\n", 764 s, __func__); 765 rstreason = BANDLIM_RST_OPENPORT; 766 goto dropwithreset; 767 } 768 } 769 #endif 770 /* 771 * Basic sanity checks on incoming SYN requests: 772 * Don't respond if the destination is a link layer 773 * broadcast according to RFC1122 4.2.3.10, p. 104. 774 * If it is from this socket it must be forged. 775 * Don't respond if the source or destination is a 776 * global or subnet broad- or multicast address. 777 * Note that it is quite possible to receive unicast 778 * link-layer packets with a broadcast IP address. Use 779 * in_broadcast() to find them. 780 */ 781 if (m->m_flags & (M_BCAST|M_MCAST)) { 782 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 783 log(LOG_DEBUG, "%s; %s: Listen socket: " 784 "Connection attempt from broad- or multicast " 785 "link layer address rejected\n", s, __func__); 786 goto dropunlock; 787 } 788 if (isipv6) { 789 #ifdef INET6 790 if (th->th_dport == th->th_sport && 791 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 792 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 793 log(LOG_DEBUG, "%s; %s: Listen socket: " 794 "Connection attempt to/from self " 795 "rejected\n", s, __func__); 796 goto dropunlock; 797 } 798 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 799 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 800 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 801 log(LOG_DEBUG, "%s; %s: Listen socket: " 802 "Connection attempt from/to multicast " 803 "address rejected\n", s, __func__); 804 goto dropunlock; 805 } 806 #endif 807 } else { 808 if (th->th_dport == th->th_sport && 809 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 810 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 811 log(LOG_DEBUG, "%s; %s: Listen socket: " 812 "Connection attempt from/to self " 813 "rejected\n", s, __func__); 814 goto dropunlock; 815 } 816 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 817 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 818 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 819 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 820 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 821 log(LOG_DEBUG, "%s; %s: Listen socket: " 822 "Connection attempt from/to broad- " 823 "or multicast address rejected\n", 824 s, __func__); 825 goto dropunlock; 826 } 827 } 828 /* 829 * SYN appears to be valid. Create compressed TCP state 830 * for syncache. 831 */ 832 #ifdef TCPDEBUG 833 if (so->so_options & SO_DEBUG) 834 tcp_trace(TA_INPUT, ostate, tp, 835 (void *)tcp_saveipgen, &tcp_savetcp, 0); 836 #endif 837 tcp_dooptions(&to, optp, optlen, TO_SYN); 838 syncache_add(&inc, &to, th, inp, &so, m); 839 /* 840 * Entry added to syncache and mbuf consumed. 841 * Everything already unlocked by syncache_add(). 842 */ 843 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 844 return; 845 } 846 847 /* 848 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 849 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 850 * the inpcb, and unlocks pcbinfo. 851 */ 852 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen); 853 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 854 return; 855 856 dropwithreset: 857 INP_INFO_WLOCK_ASSERT(&tcbinfo); 858 tcp_dropwithreset(m, th, tp, tlen, rstreason); 859 m = NULL; /* mbuf chain got consumed. */ 860 dropunlock: 861 INP_INFO_WLOCK_ASSERT(&tcbinfo); 862 if (inp != NULL) 863 INP_UNLOCK(inp); 864 INP_INFO_WUNLOCK(&tcbinfo); 865 drop: 866 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 867 if (s != NULL) 868 free(s, M_TCPLOG); 869 if (m != NULL) 870 m_freem(m); 871 return; 872 } 873 874 static void 875 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 876 struct tcpcb *tp, int drop_hdrlen, int tlen) 877 { 878 int thflags, acked, ourfinisacked, needoutput = 0; 879 int headlocked = 1; 880 int rstreason, todrop, win; 881 u_long tiwin; 882 struct tcpopt to; 883 884 #ifdef TCPDEBUG 885 /* 886 * The size of tcp_saveipgen must be the size of the max ip header, 887 * now IPv6. 888 */ 889 u_char tcp_saveipgen[IP6_HDR_LEN]; 890 struct tcphdr tcp_savetcp; 891 short ostate = 0; 892 #endif 893 thflags = th->th_flags; 894 895 INP_INFO_WLOCK_ASSERT(&tcbinfo); 896 INP_LOCK_ASSERT(tp->t_inpcb); 897 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 898 __func__)); 899 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 900 __func__)); 901 902 /* 903 * Segment received on connection. 904 * Reset idle time and keep-alive timer. 905 * XXX: This should be done after segment 906 * validation to ignore broken/spoofed segs. 907 */ 908 tp->t_rcvtime = ticks; 909 if (TCPS_HAVEESTABLISHED(tp->t_state)) 910 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 911 912 /* 913 * Unscale the window into a 32-bit value. 914 * For the SYN_SENT state the scale is zero. 915 */ 916 tiwin = th->th_win << tp->snd_scale; 917 918 /* 919 * Parse options on any incoming segment. 920 */ 921 tcp_dooptions(&to, (u_char *)(th + 1), 922 (th->th_off << 2) - sizeof(struct tcphdr), 923 (thflags & TH_SYN) ? TO_SYN : 0); 924 925 /* 926 * If echoed timestamp is later than the current time, 927 * fall back to non RFC1323 RTT calculation. Normalize 928 * timestamp if syncookies were used when this connection 929 * was established. 930 */ 931 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 932 to.to_tsecr -= tp->ts_offset; 933 if (TSTMP_GT(to.to_tsecr, ticks)) 934 to.to_tsecr = 0; 935 } 936 937 /* 938 * Process options only when we get SYN/ACK back. The SYN case 939 * for incoming connections is handled in tcp_syncache. 940 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 941 * or <SYN,ACK>) segment itself is never scaled. 942 * XXX this is traditional behavior, may need to be cleaned up. 943 */ 944 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 945 if ((to.to_flags & TOF_SCALE) && 946 (tp->t_flags & TF_REQ_SCALE)) { 947 tp->t_flags |= TF_RCVD_SCALE; 948 tp->snd_scale = to.to_wscale; 949 } 950 /* 951 * Initial send window. It will be updated with 952 * the next incoming segment to the scaled value. 953 */ 954 tp->snd_wnd = th->th_win; 955 if (to.to_flags & TOF_TS) { 956 tp->t_flags |= TF_RCVD_TSTMP; 957 tp->ts_recent = to.to_tsval; 958 tp->ts_recent_age = ticks; 959 } 960 if (to.to_flags & TOF_MSS) 961 tcp_mss(tp, to.to_mss); 962 if ((tp->t_flags & TF_SACK_PERMIT) && 963 (to.to_flags & TOF_SACKPERM) == 0) 964 tp->t_flags &= ~TF_SACK_PERMIT; 965 } 966 967 /* 968 * Header prediction: check for the two common cases 969 * of a uni-directional data xfer. If the packet has 970 * no control flags, is in-sequence, the window didn't 971 * change and we're not retransmitting, it's a 972 * candidate. If the length is zero and the ack moved 973 * forward, we're the sender side of the xfer. Just 974 * free the data acked & wake any higher level process 975 * that was blocked waiting for space. If the length 976 * is non-zero and the ack didn't move, we're the 977 * receiver side. If we're getting packets in-order 978 * (the reassembly queue is empty), add the data to 979 * the socket buffer and note that we need a delayed ack. 980 * Make sure that the hidden state-flags are also off. 981 * Since we check for TCPS_ESTABLISHED first, it can only 982 * be TH_NEEDSYN. 983 */ 984 if (tp->t_state == TCPS_ESTABLISHED && 985 th->th_seq == tp->rcv_nxt && 986 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 987 tp->snd_nxt == tp->snd_max && 988 tiwin && tiwin == tp->snd_wnd && 989 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 990 LIST_EMPTY(&tp->t_segq) && 991 ((to.to_flags & TOF_TS) == 0 || 992 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 993 994 /* 995 * If last ACK falls within this segment's sequence numbers, 996 * record the timestamp. 997 * NOTE that the test is modified according to the latest 998 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 999 */ 1000 if ((to.to_flags & TOF_TS) != 0 && 1001 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1002 tp->ts_recent_age = ticks; 1003 tp->ts_recent = to.to_tsval; 1004 } 1005 1006 if (tlen == 0) { 1007 if (SEQ_GT(th->th_ack, tp->snd_una) && 1008 SEQ_LEQ(th->th_ack, tp->snd_max) && 1009 tp->snd_cwnd >= tp->snd_wnd && 1010 ((!tcp_do_newreno && 1011 !(tp->t_flags & TF_SACK_PERMIT) && 1012 tp->t_dupacks < tcprexmtthresh) || 1013 ((tcp_do_newreno || 1014 (tp->t_flags & TF_SACK_PERMIT)) && 1015 !IN_FASTRECOVERY(tp) && 1016 (to.to_flags & TOF_SACK) == 0 && 1017 TAILQ_EMPTY(&tp->snd_holes)))) { 1018 KASSERT(headlocked, 1019 ("%s: headlocked", __func__)); 1020 INP_INFO_WUNLOCK(&tcbinfo); 1021 headlocked = 0; 1022 /* 1023 * This is a pure ack for outstanding data. 1024 */ 1025 ++tcpstat.tcps_predack; 1026 /* 1027 * "bad retransmit" recovery. 1028 */ 1029 if (tp->t_rxtshift == 1 && 1030 ticks < tp->t_badrxtwin) { 1031 ++tcpstat.tcps_sndrexmitbad; 1032 tp->snd_cwnd = tp->snd_cwnd_prev; 1033 tp->snd_ssthresh = 1034 tp->snd_ssthresh_prev; 1035 tp->snd_recover = tp->snd_recover_prev; 1036 if (tp->t_flags & TF_WASFRECOVERY) 1037 ENTER_FASTRECOVERY(tp); 1038 tp->snd_nxt = tp->snd_max; 1039 tp->t_badrxtwin = 0; 1040 } 1041 1042 /* 1043 * Recalculate the transmit timer / rtt. 1044 * 1045 * Some boxes send broken timestamp replies 1046 * during the SYN+ACK phase, ignore 1047 * timestamps of 0 or we could calculate a 1048 * huge RTT and blow up the retransmit timer. 1049 */ 1050 if ((to.to_flags & TOF_TS) != 0 && 1051 to.to_tsecr) { 1052 if (!tp->t_rttlow || 1053 tp->t_rttlow > ticks - to.to_tsecr) 1054 tp->t_rttlow = ticks - to.to_tsecr; 1055 tcp_xmit_timer(tp, 1056 ticks - to.to_tsecr + 1); 1057 } else if (tp->t_rtttime && 1058 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1059 if (!tp->t_rttlow || 1060 tp->t_rttlow > ticks - tp->t_rtttime) 1061 tp->t_rttlow = ticks - tp->t_rtttime; 1062 tcp_xmit_timer(tp, 1063 ticks - tp->t_rtttime); 1064 } 1065 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1066 acked = th->th_ack - tp->snd_una; 1067 tcpstat.tcps_rcvackpack++; 1068 tcpstat.tcps_rcvackbyte += acked; 1069 sbdrop(&so->so_snd, acked); 1070 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1071 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1072 tp->snd_recover = th->th_ack - 1; 1073 tp->snd_una = th->th_ack; 1074 /* 1075 * Pull snd_wl2 up to prevent seq wrap relative 1076 * to th_ack. 1077 */ 1078 tp->snd_wl2 = th->th_ack; 1079 tp->t_dupacks = 0; 1080 m_freem(m); 1081 ND6_HINT(tp); /* Some progress has been made. */ 1082 1083 /* 1084 * If all outstanding data are acked, stop 1085 * retransmit timer, otherwise restart timer 1086 * using current (possibly backed-off) value. 1087 * If process is waiting for space, 1088 * wakeup/selwakeup/signal. If data 1089 * are ready to send, let tcp_output 1090 * decide between more output or persist. 1091 */ 1092 #ifdef TCPDEBUG 1093 if (so->so_options & SO_DEBUG) 1094 tcp_trace(TA_INPUT, ostate, tp, 1095 (void *)tcp_saveipgen, 1096 &tcp_savetcp, 0); 1097 #endif 1098 if (tp->snd_una == tp->snd_max) 1099 tcp_timer_activate(tp, TT_REXMT, 0); 1100 else if (!tcp_timer_active(tp, TT_PERSIST)) 1101 tcp_timer_activate(tp, TT_REXMT, 1102 tp->t_rxtcur); 1103 /* 1104 * NB: sowwakeup_locked() does an 1105 * implicit unlock. 1106 */ 1107 sowwakeup(so); 1108 if (so->so_snd.sb_cc) 1109 (void) tcp_output(tp); 1110 goto check_delack; 1111 } 1112 } else if (th->th_ack == tp->snd_una && 1113 tlen <= sbspace(&so->so_rcv)) { 1114 int newsize = 0; /* automatic sockbuf scaling */ 1115 1116 KASSERT(headlocked, ("%s: headlocked", __func__)); 1117 INP_INFO_WUNLOCK(&tcbinfo); 1118 headlocked = 0; 1119 /* 1120 * This is a pure, in-sequence data packet 1121 * with nothing on the reassembly queue and 1122 * we have enough buffer space to take it. 1123 */ 1124 /* Clean receiver SACK report if present */ 1125 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1126 tcp_clean_sackreport(tp); 1127 ++tcpstat.tcps_preddat; 1128 tp->rcv_nxt += tlen; 1129 /* 1130 * Pull snd_wl1 up to prevent seq wrap relative to 1131 * th_seq. 1132 */ 1133 tp->snd_wl1 = th->th_seq; 1134 /* 1135 * Pull rcv_up up to prevent seq wrap relative to 1136 * rcv_nxt. 1137 */ 1138 tp->rcv_up = tp->rcv_nxt; 1139 tcpstat.tcps_rcvpack++; 1140 tcpstat.tcps_rcvbyte += tlen; 1141 ND6_HINT(tp); /* Some progress has been made */ 1142 #ifdef TCPDEBUG 1143 if (so->so_options & SO_DEBUG) 1144 tcp_trace(TA_INPUT, ostate, tp, 1145 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1146 #endif 1147 /* 1148 * Automatic sizing of receive socket buffer. Often the send 1149 * buffer size is not optimally adjusted to the actual network 1150 * conditions at hand (delay bandwidth product). Setting the 1151 * buffer size too small limits throughput on links with high 1152 * bandwidth and high delay (eg. trans-continental/oceanic links). 1153 * 1154 * On the receive side the socket buffer memory is only rarely 1155 * used to any significant extent. This allows us to be much 1156 * more aggressive in scaling the receive socket buffer. For 1157 * the case that the buffer space is actually used to a large 1158 * extent and we run out of kernel memory we can simply drop 1159 * the new segments; TCP on the sender will just retransmit it 1160 * later. Setting the buffer size too big may only consume too 1161 * much kernel memory if the application doesn't read() from 1162 * the socket or packet loss or reordering makes use of the 1163 * reassembly queue. 1164 * 1165 * The criteria to step up the receive buffer one notch are: 1166 * 1. the number of bytes received during the time it takes 1167 * one timestamp to be reflected back to us (the RTT); 1168 * 2. received bytes per RTT is within seven eighth of the 1169 * current socket buffer size; 1170 * 3. receive buffer size has not hit maximal automatic size; 1171 * 1172 * This algorithm does one step per RTT at most and only if 1173 * we receive a bulk stream w/o packet losses or reorderings. 1174 * Shrinking the buffer during idle times is not necessary as 1175 * it doesn't consume any memory when idle. 1176 * 1177 * TODO: Only step up if the application is actually serving 1178 * the buffer to better manage the socket buffer resources. 1179 */ 1180 if (tcp_do_autorcvbuf && 1181 to.to_tsecr && 1182 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1183 if (to.to_tsecr > tp->rfbuf_ts && 1184 to.to_tsecr - tp->rfbuf_ts < hz) { 1185 if (tp->rfbuf_cnt > 1186 (so->so_rcv.sb_hiwat / 8 * 7) && 1187 so->so_rcv.sb_hiwat < 1188 tcp_autorcvbuf_max) { 1189 newsize = 1190 min(so->so_rcv.sb_hiwat + 1191 tcp_autorcvbuf_inc, 1192 tcp_autorcvbuf_max); 1193 } 1194 /* Start over with next RTT. */ 1195 tp->rfbuf_ts = 0; 1196 tp->rfbuf_cnt = 0; 1197 } else 1198 tp->rfbuf_cnt += tlen; /* add up */ 1199 } 1200 1201 /* Add data to socket buffer. */ 1202 SOCKBUF_LOCK(&so->so_rcv); 1203 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1204 m_freem(m); 1205 } else { 1206 /* 1207 * Set new socket buffer size. 1208 * Give up when limit is reached. 1209 */ 1210 if (newsize) 1211 if (!sbreserve_locked(&so->so_rcv, 1212 newsize, so, curthread)) 1213 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1214 m_adj(m, drop_hdrlen); /* delayed header drop */ 1215 sbappendstream_locked(&so->so_rcv, m); 1216 } 1217 /* NB: sorwakeup_locked() does an implicit unlock. */ 1218 sorwakeup_locked(so); 1219 if (DELAY_ACK(tp)) { 1220 tp->t_flags |= TF_DELACK; 1221 } else { 1222 tp->t_flags |= TF_ACKNOW; 1223 tcp_output(tp); 1224 } 1225 goto check_delack; 1226 } 1227 } 1228 1229 /* 1230 * Calculate amount of space in receive window, 1231 * and then do TCP input processing. 1232 * Receive window is amount of space in rcv queue, 1233 * but not less than advertised window. 1234 */ 1235 win = sbspace(&so->so_rcv); 1236 if (win < 0) 1237 win = 0; 1238 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1239 1240 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1241 tp->rfbuf_ts = 0; 1242 tp->rfbuf_cnt = 0; 1243 1244 switch (tp->t_state) { 1245 1246 /* 1247 * If the state is SYN_RECEIVED: 1248 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1249 */ 1250 case TCPS_SYN_RECEIVED: 1251 if ((thflags & TH_ACK) && 1252 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1253 SEQ_GT(th->th_ack, tp->snd_max))) { 1254 rstreason = BANDLIM_RST_OPENPORT; 1255 goto dropwithreset; 1256 } 1257 break; 1258 1259 /* 1260 * If the state is SYN_SENT: 1261 * if seg contains an ACK, but not for our SYN, drop the input. 1262 * if seg contains a RST, then drop the connection. 1263 * if seg does not contain SYN, then drop it. 1264 * Otherwise this is an acceptable SYN segment 1265 * initialize tp->rcv_nxt and tp->irs 1266 * if seg contains ack then advance tp->snd_una 1267 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1268 * arrange for segment to be acked (eventually) 1269 * continue processing rest of data/controls, beginning with URG 1270 */ 1271 case TCPS_SYN_SENT: 1272 if ((thflags & TH_ACK) && 1273 (SEQ_LEQ(th->th_ack, tp->iss) || 1274 SEQ_GT(th->th_ack, tp->snd_max))) { 1275 rstreason = BANDLIM_UNLIMITED; 1276 goto dropwithreset; 1277 } 1278 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1279 tp = tcp_drop(tp, ECONNREFUSED); 1280 if (thflags & TH_RST) 1281 goto drop; 1282 if (!(thflags & TH_SYN)) 1283 goto drop; 1284 1285 tp->irs = th->th_seq; 1286 tcp_rcvseqinit(tp); 1287 if (thflags & TH_ACK) { 1288 tcpstat.tcps_connects++; 1289 soisconnected(so); 1290 #ifdef MAC 1291 SOCK_LOCK(so); 1292 mac_set_socket_peer_from_mbuf(m, so); 1293 SOCK_UNLOCK(so); 1294 #endif 1295 /* Do window scaling on this connection? */ 1296 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1297 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1298 tp->rcv_scale = tp->request_r_scale; 1299 } 1300 tp->rcv_adv += tp->rcv_wnd; 1301 tp->snd_una++; /* SYN is acked */ 1302 /* 1303 * If there's data, delay ACK; if there's also a FIN 1304 * ACKNOW will be turned on later. 1305 */ 1306 if (DELAY_ACK(tp) && tlen != 0) 1307 tcp_timer_activate(tp, TT_DELACK, 1308 tcp_delacktime); 1309 else 1310 tp->t_flags |= TF_ACKNOW; 1311 /* 1312 * Received <SYN,ACK> in SYN_SENT[*] state. 1313 * Transitions: 1314 * SYN_SENT --> ESTABLISHED 1315 * SYN_SENT* --> FIN_WAIT_1 1316 */ 1317 tp->t_starttime = ticks; 1318 if (tp->t_flags & TF_NEEDFIN) { 1319 tp->t_state = TCPS_FIN_WAIT_1; 1320 tp->t_flags &= ~TF_NEEDFIN; 1321 thflags &= ~TH_SYN; 1322 } else { 1323 tp->t_state = TCPS_ESTABLISHED; 1324 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1325 } 1326 } else { 1327 /* 1328 * Received initial SYN in SYN-SENT[*] state => 1329 * simultaneous open. If segment contains CC option 1330 * and there is a cached CC, apply TAO test. 1331 * If it succeeds, connection is * half-synchronized. 1332 * Otherwise, do 3-way handshake: 1333 * SYN-SENT -> SYN-RECEIVED 1334 * SYN-SENT* -> SYN-RECEIVED* 1335 * If there was no CC option, clear cached CC value. 1336 */ 1337 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1338 tcp_timer_activate(tp, TT_REXMT, 0); 1339 tp->t_state = TCPS_SYN_RECEIVED; 1340 } 1341 1342 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1343 __func__)); 1344 INP_LOCK_ASSERT(tp->t_inpcb); 1345 1346 /* 1347 * Advance th->th_seq to correspond to first data byte. 1348 * If data, trim to stay within window, 1349 * dropping FIN if necessary. 1350 */ 1351 th->th_seq++; 1352 if (tlen > tp->rcv_wnd) { 1353 todrop = tlen - tp->rcv_wnd; 1354 m_adj(m, -todrop); 1355 tlen = tp->rcv_wnd; 1356 thflags &= ~TH_FIN; 1357 tcpstat.tcps_rcvpackafterwin++; 1358 tcpstat.tcps_rcvbyteafterwin += todrop; 1359 } 1360 tp->snd_wl1 = th->th_seq - 1; 1361 tp->rcv_up = th->th_seq; 1362 /* 1363 * Client side of transaction: already sent SYN and data. 1364 * If the remote host used T/TCP to validate the SYN, 1365 * our data will be ACK'd; if so, enter normal data segment 1366 * processing in the middle of step 5, ack processing. 1367 * Otherwise, goto step 6. 1368 */ 1369 if (thflags & TH_ACK) 1370 goto process_ACK; 1371 1372 goto step6; 1373 1374 /* 1375 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1376 * do normal processing. 1377 * 1378 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1379 */ 1380 case TCPS_LAST_ACK: 1381 case TCPS_CLOSING: 1382 break; /* continue normal processing */ 1383 } 1384 1385 /* 1386 * States other than LISTEN or SYN_SENT. 1387 * First check the RST flag and sequence number since reset segments 1388 * are exempt from the timestamp and connection count tests. This 1389 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1390 * below which allowed reset segments in half the sequence space 1391 * to fall though and be processed (which gives forged reset 1392 * segments with a random sequence number a 50 percent chance of 1393 * killing a connection). 1394 * Then check timestamp, if present. 1395 * Then check the connection count, if present. 1396 * Then check that at least some bytes of segment are within 1397 * receive window. If segment begins before rcv_nxt, 1398 * drop leading data (and SYN); if nothing left, just ack. 1399 * 1400 * 1401 * If the RST bit is set, check the sequence number to see 1402 * if this is a valid reset segment. 1403 * RFC 793 page 37: 1404 * In all states except SYN-SENT, all reset (RST) segments 1405 * are validated by checking their SEQ-fields. A reset is 1406 * valid if its sequence number is in the window. 1407 * Note: this does not take into account delayed ACKs, so 1408 * we should test against last_ack_sent instead of rcv_nxt. 1409 * The sequence number in the reset segment is normally an 1410 * echo of our outgoing acknowlegement numbers, but some hosts 1411 * send a reset with the sequence number at the rightmost edge 1412 * of our receive window, and we have to handle this case. 1413 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1414 * that brute force RST attacks are possible. To combat this, 1415 * we use a much stricter check while in the ESTABLISHED state, 1416 * only accepting RSTs where the sequence number is equal to 1417 * last_ack_sent. In all other states (the states in which a 1418 * RST is more likely), the more permissive check is used. 1419 * If we have multiple segments in flight, the intial reset 1420 * segment sequence numbers will be to the left of last_ack_sent, 1421 * but they will eventually catch up. 1422 * In any case, it never made sense to trim reset segments to 1423 * fit the receive window since RFC 1122 says: 1424 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1425 * 1426 * A TCP SHOULD allow a received RST segment to include data. 1427 * 1428 * DISCUSSION 1429 * It has been suggested that a RST segment could contain 1430 * ASCII text that encoded and explained the cause of the 1431 * RST. No standard has yet been established for such 1432 * data. 1433 * 1434 * If the reset segment passes the sequence number test examine 1435 * the state: 1436 * SYN_RECEIVED STATE: 1437 * If passive open, return to LISTEN state. 1438 * If active open, inform user that connection was refused. 1439 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1440 * Inform user that connection was reset, and close tcb. 1441 * CLOSING, LAST_ACK STATES: 1442 * Close the tcb. 1443 * TIME_WAIT STATE: 1444 * Drop the segment - see Stevens, vol. 2, p. 964 and 1445 * RFC 1337. 1446 */ 1447 if (thflags & TH_RST) { 1448 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1449 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1450 switch (tp->t_state) { 1451 1452 case TCPS_SYN_RECEIVED: 1453 so->so_error = ECONNREFUSED; 1454 goto close; 1455 1456 case TCPS_ESTABLISHED: 1457 if (tcp_insecure_rst == 0 && 1458 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1459 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1460 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1461 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1462 tcpstat.tcps_badrst++; 1463 goto drop; 1464 } 1465 case TCPS_FIN_WAIT_1: 1466 case TCPS_FIN_WAIT_2: 1467 case TCPS_CLOSE_WAIT: 1468 so->so_error = ECONNRESET; 1469 close: 1470 tp->t_state = TCPS_CLOSED; 1471 tcpstat.tcps_drops++; 1472 KASSERT(headlocked, ("%s: trimthenstep6: " 1473 "tcp_close: head not locked", __func__)); 1474 tp = tcp_close(tp); 1475 break; 1476 1477 case TCPS_CLOSING: 1478 case TCPS_LAST_ACK: 1479 KASSERT(headlocked, ("%s: trimthenstep6: " 1480 "tcp_close.2: head not locked", __func__)); 1481 tp = tcp_close(tp); 1482 break; 1483 } 1484 } 1485 goto drop; 1486 } 1487 1488 /* 1489 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1490 * and it's less than ts_recent, drop it. 1491 */ 1492 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1493 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1494 1495 /* Check to see if ts_recent is over 24 days old. */ 1496 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1497 /* 1498 * Invalidate ts_recent. If this segment updates 1499 * ts_recent, the age will be reset later and ts_recent 1500 * will get a valid value. If it does not, setting 1501 * ts_recent to zero will at least satisfy the 1502 * requirement that zero be placed in the timestamp 1503 * echo reply when ts_recent isn't valid. The 1504 * age isn't reset until we get a valid ts_recent 1505 * because we don't want out-of-order segments to be 1506 * dropped when ts_recent is old. 1507 */ 1508 tp->ts_recent = 0; 1509 } else { 1510 tcpstat.tcps_rcvduppack++; 1511 tcpstat.tcps_rcvdupbyte += tlen; 1512 tcpstat.tcps_pawsdrop++; 1513 if (tlen) 1514 goto dropafterack; 1515 goto drop; 1516 } 1517 } 1518 1519 /* 1520 * In the SYN-RECEIVED state, validate that the packet belongs to 1521 * this connection before trimming the data to fit the receive 1522 * window. Check the sequence number versus IRS since we know 1523 * the sequence numbers haven't wrapped. This is a partial fix 1524 * for the "LAND" DoS attack. 1525 */ 1526 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1527 rstreason = BANDLIM_RST_OPENPORT; 1528 goto dropwithreset; 1529 } 1530 1531 todrop = tp->rcv_nxt - th->th_seq; 1532 if (todrop > 0) { 1533 if (thflags & TH_SYN) { 1534 thflags &= ~TH_SYN; 1535 th->th_seq++; 1536 if (th->th_urp > 1) 1537 th->th_urp--; 1538 else 1539 thflags &= ~TH_URG; 1540 todrop--; 1541 } 1542 /* 1543 * Following if statement from Stevens, vol. 2, p. 960. 1544 */ 1545 if (todrop > tlen 1546 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1547 /* 1548 * Any valid FIN must be to the left of the window. 1549 * At this point the FIN must be a duplicate or out 1550 * of sequence; drop it. 1551 */ 1552 thflags &= ~TH_FIN; 1553 1554 /* 1555 * Send an ACK to resynchronize and drop any data. 1556 * But keep on processing for RST or ACK. 1557 */ 1558 tp->t_flags |= TF_ACKNOW; 1559 todrop = tlen; 1560 tcpstat.tcps_rcvduppack++; 1561 tcpstat.tcps_rcvdupbyte += todrop; 1562 } else { 1563 tcpstat.tcps_rcvpartduppack++; 1564 tcpstat.tcps_rcvpartdupbyte += todrop; 1565 } 1566 drop_hdrlen += todrop; /* drop from the top afterwards */ 1567 th->th_seq += todrop; 1568 tlen -= todrop; 1569 if (th->th_urp > todrop) 1570 th->th_urp -= todrop; 1571 else { 1572 thflags &= ~TH_URG; 1573 th->th_urp = 0; 1574 } 1575 } 1576 1577 /* 1578 * If new data are received on a connection after the 1579 * user processes are gone, then RST the other end. 1580 */ 1581 if ((so->so_state & SS_NOFDREF) && 1582 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1583 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1584 "not locked", __func__)); 1585 tp = tcp_close(tp); 1586 tcpstat.tcps_rcvafterclose++; 1587 rstreason = BANDLIM_UNLIMITED; 1588 goto dropwithreset; 1589 } 1590 1591 /* 1592 * If segment ends after window, drop trailing data 1593 * (and PUSH and FIN); if nothing left, just ACK. 1594 */ 1595 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1596 if (todrop > 0) { 1597 tcpstat.tcps_rcvpackafterwin++; 1598 if (todrop >= tlen) { 1599 tcpstat.tcps_rcvbyteafterwin += tlen; 1600 /* 1601 * If window is closed can only take segments at 1602 * window edge, and have to drop data and PUSH from 1603 * incoming segments. Continue processing, but 1604 * remember to ack. Otherwise, drop segment 1605 * and ack. 1606 */ 1607 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1608 tp->t_flags |= TF_ACKNOW; 1609 tcpstat.tcps_rcvwinprobe++; 1610 } else 1611 goto dropafterack; 1612 } else 1613 tcpstat.tcps_rcvbyteafterwin += todrop; 1614 m_adj(m, -todrop); 1615 tlen -= todrop; 1616 thflags &= ~(TH_PUSH|TH_FIN); 1617 } 1618 1619 /* 1620 * If last ACK falls within this segment's sequence numbers, 1621 * record its timestamp. 1622 * NOTE: 1623 * 1) That the test incorporates suggestions from the latest 1624 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1625 * 2) That updating only on newer timestamps interferes with 1626 * our earlier PAWS tests, so this check should be solely 1627 * predicated on the sequence space of this segment. 1628 * 3) That we modify the segment boundary check to be 1629 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1630 * instead of RFC1323's 1631 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1632 * This modified check allows us to overcome RFC1323's 1633 * limitations as described in Stevens TCP/IP Illustrated 1634 * Vol. 2 p.869. In such cases, we can still calculate the 1635 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1636 */ 1637 if ((to.to_flags & TOF_TS) != 0 && 1638 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1639 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1640 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1641 tp->ts_recent_age = ticks; 1642 tp->ts_recent = to.to_tsval; 1643 } 1644 1645 /* 1646 * If a SYN is in the window, then this is an 1647 * error and we send an RST and drop the connection. 1648 */ 1649 if (thflags & TH_SYN) { 1650 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1651 "head not locked", __func__)); 1652 tp = tcp_drop(tp, ECONNRESET); 1653 rstreason = BANDLIM_UNLIMITED; 1654 goto drop; 1655 } 1656 1657 /* 1658 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1659 * flag is on (half-synchronized state), then queue data for 1660 * later processing; else drop segment and return. 1661 */ 1662 if ((thflags & TH_ACK) == 0) { 1663 if (tp->t_state == TCPS_SYN_RECEIVED || 1664 (tp->t_flags & TF_NEEDSYN)) 1665 goto step6; 1666 else if (tp->t_flags & TF_ACKNOW) 1667 goto dropafterack; 1668 else 1669 goto drop; 1670 } 1671 1672 /* 1673 * Ack processing. 1674 */ 1675 switch (tp->t_state) { 1676 1677 /* 1678 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1679 * ESTABLISHED state and continue processing. 1680 * The ACK was checked above. 1681 */ 1682 case TCPS_SYN_RECEIVED: 1683 1684 tcpstat.tcps_connects++; 1685 soisconnected(so); 1686 /* Do window scaling? */ 1687 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1688 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1689 tp->rcv_scale = tp->request_r_scale; 1690 tp->snd_wnd = tiwin; 1691 } 1692 /* 1693 * Make transitions: 1694 * SYN-RECEIVED -> ESTABLISHED 1695 * SYN-RECEIVED* -> FIN-WAIT-1 1696 */ 1697 tp->t_starttime = ticks; 1698 if (tp->t_flags & TF_NEEDFIN) { 1699 tp->t_state = TCPS_FIN_WAIT_1; 1700 tp->t_flags &= ~TF_NEEDFIN; 1701 } else { 1702 tp->t_state = TCPS_ESTABLISHED; 1703 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1704 } 1705 /* 1706 * If segment contains data or ACK, will call tcp_reass() 1707 * later; if not, do so now to pass queued data to user. 1708 */ 1709 if (tlen == 0 && (thflags & TH_FIN) == 0) 1710 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1711 (struct mbuf *)0); 1712 tp->snd_wl1 = th->th_seq - 1; 1713 /* FALLTHROUGH */ 1714 1715 /* 1716 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1717 * ACKs. If the ack is in the range 1718 * tp->snd_una < th->th_ack <= tp->snd_max 1719 * then advance tp->snd_una to th->th_ack and drop 1720 * data from the retransmission queue. If this ACK reflects 1721 * more up to date window information we update our window information. 1722 */ 1723 case TCPS_ESTABLISHED: 1724 case TCPS_FIN_WAIT_1: 1725 case TCPS_FIN_WAIT_2: 1726 case TCPS_CLOSE_WAIT: 1727 case TCPS_CLOSING: 1728 case TCPS_LAST_ACK: 1729 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1730 tcpstat.tcps_rcvacktoomuch++; 1731 goto dropafterack; 1732 } 1733 if ((tp->t_flags & TF_SACK_PERMIT) && 1734 ((to.to_flags & TOF_SACK) || 1735 !TAILQ_EMPTY(&tp->snd_holes))) 1736 tcp_sack_doack(tp, &to, th->th_ack); 1737 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1738 if (tlen == 0 && tiwin == tp->snd_wnd) { 1739 tcpstat.tcps_rcvdupack++; 1740 /* 1741 * If we have outstanding data (other than 1742 * a window probe), this is a completely 1743 * duplicate ack (ie, window info didn't 1744 * change), the ack is the biggest we've 1745 * seen and we've seen exactly our rexmt 1746 * threshhold of them, assume a packet 1747 * has been dropped and retransmit it. 1748 * Kludge snd_nxt & the congestion 1749 * window so we send only this one 1750 * packet. 1751 * 1752 * We know we're losing at the current 1753 * window size so do congestion avoidance 1754 * (set ssthresh to half the current window 1755 * and pull our congestion window back to 1756 * the new ssthresh). 1757 * 1758 * Dup acks mean that packets have left the 1759 * network (they're now cached at the receiver) 1760 * so bump cwnd by the amount in the receiver 1761 * to keep a constant cwnd packets in the 1762 * network. 1763 */ 1764 if (!tcp_timer_active(tp, TT_REXMT) || 1765 th->th_ack != tp->snd_una) 1766 tp->t_dupacks = 0; 1767 else if (++tp->t_dupacks > tcprexmtthresh || 1768 ((tcp_do_newreno || 1769 (tp->t_flags & TF_SACK_PERMIT)) && 1770 IN_FASTRECOVERY(tp))) { 1771 if ((tp->t_flags & TF_SACK_PERMIT) && 1772 IN_FASTRECOVERY(tp)) { 1773 int awnd; 1774 1775 /* 1776 * Compute the amount of data in flight first. 1777 * We can inject new data into the pipe iff 1778 * we have less than 1/2 the original window's 1779 * worth of data in flight. 1780 */ 1781 awnd = (tp->snd_nxt - tp->snd_fack) + 1782 tp->sackhint.sack_bytes_rexmit; 1783 if (awnd < tp->snd_ssthresh) { 1784 tp->snd_cwnd += tp->t_maxseg; 1785 if (tp->snd_cwnd > tp->snd_ssthresh) 1786 tp->snd_cwnd = tp->snd_ssthresh; 1787 } 1788 } else 1789 tp->snd_cwnd += tp->t_maxseg; 1790 (void) tcp_output(tp); 1791 goto drop; 1792 } else if (tp->t_dupacks == tcprexmtthresh) { 1793 tcp_seq onxt = tp->snd_nxt; 1794 u_int win; 1795 1796 /* 1797 * If we're doing sack, check to 1798 * see if we're already in sack 1799 * recovery. If we're not doing sack, 1800 * check to see if we're in newreno 1801 * recovery. 1802 */ 1803 if (tp->t_flags & TF_SACK_PERMIT) { 1804 if (IN_FASTRECOVERY(tp)) { 1805 tp->t_dupacks = 0; 1806 break; 1807 } 1808 } else if (tcp_do_newreno) { 1809 if (SEQ_LEQ(th->th_ack, 1810 tp->snd_recover)) { 1811 tp->t_dupacks = 0; 1812 break; 1813 } 1814 } 1815 win = min(tp->snd_wnd, tp->snd_cwnd) / 1816 2 / tp->t_maxseg; 1817 if (win < 2) 1818 win = 2; 1819 tp->snd_ssthresh = win * tp->t_maxseg; 1820 ENTER_FASTRECOVERY(tp); 1821 tp->snd_recover = tp->snd_max; 1822 tcp_timer_activate(tp, TT_REXMT, 0); 1823 tp->t_rtttime = 0; 1824 if (tp->t_flags & TF_SACK_PERMIT) { 1825 tcpstat.tcps_sack_recovery_episode++; 1826 tp->sack_newdata = tp->snd_nxt; 1827 tp->snd_cwnd = tp->t_maxseg; 1828 (void) tcp_output(tp); 1829 goto drop; 1830 } 1831 tp->snd_nxt = th->th_ack; 1832 tp->snd_cwnd = tp->t_maxseg; 1833 (void) tcp_output(tp); 1834 KASSERT(tp->snd_limited <= 2, 1835 ("%s: tp->snd_limited too big", 1836 __func__)); 1837 tp->snd_cwnd = tp->snd_ssthresh + 1838 tp->t_maxseg * 1839 (tp->t_dupacks - tp->snd_limited); 1840 if (SEQ_GT(onxt, tp->snd_nxt)) 1841 tp->snd_nxt = onxt; 1842 goto drop; 1843 } else if (tcp_do_rfc3042) { 1844 u_long oldcwnd = tp->snd_cwnd; 1845 tcp_seq oldsndmax = tp->snd_max; 1846 u_int sent; 1847 1848 KASSERT(tp->t_dupacks == 1 || 1849 tp->t_dupacks == 2, 1850 ("%s: dupacks not 1 or 2", 1851 __func__)); 1852 if (tp->t_dupacks == 1) 1853 tp->snd_limited = 0; 1854 tp->snd_cwnd = 1855 (tp->snd_nxt - tp->snd_una) + 1856 (tp->t_dupacks - tp->snd_limited) * 1857 tp->t_maxseg; 1858 (void) tcp_output(tp); 1859 sent = tp->snd_max - oldsndmax; 1860 if (sent > tp->t_maxseg) { 1861 KASSERT((tp->t_dupacks == 2 && 1862 tp->snd_limited == 0) || 1863 (sent == tp->t_maxseg + 1 && 1864 tp->t_flags & TF_SENTFIN), 1865 ("%s: sent too much", 1866 __func__)); 1867 tp->snd_limited = 2; 1868 } else if (sent > 0) 1869 ++tp->snd_limited; 1870 tp->snd_cwnd = oldcwnd; 1871 goto drop; 1872 } 1873 } else 1874 tp->t_dupacks = 0; 1875 break; 1876 } 1877 1878 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 1879 ("%s: th_ack <= snd_una", __func__)); 1880 1881 /* 1882 * If the congestion window was inflated to account 1883 * for the other side's cached packets, retract it. 1884 */ 1885 if (tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) { 1886 if (IN_FASTRECOVERY(tp)) { 1887 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1888 if (tp->t_flags & TF_SACK_PERMIT) 1889 tcp_sack_partialack(tp, th); 1890 else 1891 tcp_newreno_partial_ack(tp, th); 1892 } else { 1893 /* 1894 * Out of fast recovery. 1895 * Window inflation should have left us 1896 * with approximately snd_ssthresh 1897 * outstanding data. 1898 * But in case we would be inclined to 1899 * send a burst, better to do it via 1900 * the slow start mechanism. 1901 */ 1902 if (SEQ_GT(th->th_ack + 1903 tp->snd_ssthresh, 1904 tp->snd_max)) 1905 tp->snd_cwnd = tp->snd_max - 1906 th->th_ack + 1907 tp->t_maxseg; 1908 else 1909 tp->snd_cwnd = tp->snd_ssthresh; 1910 } 1911 } 1912 } else { 1913 if (tp->t_dupacks >= tcprexmtthresh && 1914 tp->snd_cwnd > tp->snd_ssthresh) 1915 tp->snd_cwnd = tp->snd_ssthresh; 1916 } 1917 tp->t_dupacks = 0; 1918 /* 1919 * If we reach this point, ACK is not a duplicate, 1920 * i.e., it ACKs something we sent. 1921 */ 1922 if (tp->t_flags & TF_NEEDSYN) { 1923 /* 1924 * T/TCP: Connection was half-synchronized, and our 1925 * SYN has been ACK'd (so connection is now fully 1926 * synchronized). Go to non-starred state, 1927 * increment snd_una for ACK of SYN, and check if 1928 * we can do window scaling. 1929 */ 1930 tp->t_flags &= ~TF_NEEDSYN; 1931 tp->snd_una++; 1932 /* Do window scaling? */ 1933 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1934 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1935 tp->rcv_scale = tp->request_r_scale; 1936 /* Send window already scaled. */ 1937 } 1938 } 1939 1940 process_ACK: 1941 KASSERT(headlocked, ("%s: process_ACK: head not locked", 1942 __func__)); 1943 INP_LOCK_ASSERT(tp->t_inpcb); 1944 1945 acked = th->th_ack - tp->snd_una; 1946 tcpstat.tcps_rcvackpack++; 1947 tcpstat.tcps_rcvackbyte += acked; 1948 1949 /* 1950 * If we just performed our first retransmit, and the ACK 1951 * arrives within our recovery window, then it was a mistake 1952 * to do the retransmit in the first place. Recover our 1953 * original cwnd and ssthresh, and proceed to transmit where 1954 * we left off. 1955 */ 1956 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1957 ++tcpstat.tcps_sndrexmitbad; 1958 tp->snd_cwnd = tp->snd_cwnd_prev; 1959 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1960 tp->snd_recover = tp->snd_recover_prev; 1961 if (tp->t_flags & TF_WASFRECOVERY) 1962 ENTER_FASTRECOVERY(tp); 1963 tp->snd_nxt = tp->snd_max; 1964 tp->t_badrxtwin = 0; /* XXX probably not required */ 1965 } 1966 1967 /* 1968 * If we have a timestamp reply, update smoothed 1969 * round trip time. If no timestamp is present but 1970 * transmit timer is running and timed sequence 1971 * number was acked, update smoothed round trip time. 1972 * Since we now have an rtt measurement, cancel the 1973 * timer backoff (cf., Phil Karn's retransmit alg.). 1974 * Recompute the initial retransmit timer. 1975 * 1976 * Some boxes send broken timestamp replies 1977 * during the SYN+ACK phase, ignore 1978 * timestamps of 0 or we could calculate a 1979 * huge RTT and blow up the retransmit timer. 1980 */ 1981 if ((to.to_flags & TOF_TS) != 0 && 1982 to.to_tsecr) { 1983 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 1984 tp->t_rttlow = ticks - to.to_tsecr; 1985 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1986 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1987 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 1988 tp->t_rttlow = ticks - tp->t_rtttime; 1989 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1990 } 1991 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1992 1993 /* 1994 * If all outstanding data is acked, stop retransmit 1995 * timer and remember to restart (more output or persist). 1996 * If there is more data to be acked, restart retransmit 1997 * timer, using current (possibly backed-off) value. 1998 */ 1999 if (th->th_ack == tp->snd_max) { 2000 tcp_timer_activate(tp, TT_REXMT, 0); 2001 needoutput = 1; 2002 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2003 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2004 2005 /* 2006 * If no data (only SYN) was ACK'd, 2007 * skip rest of ACK processing. 2008 */ 2009 if (acked == 0) 2010 goto step6; 2011 2012 /* 2013 * When new data is acked, open the congestion window. 2014 * If the window gives us less than ssthresh packets 2015 * in flight, open exponentially (maxseg per packet). 2016 * Otherwise open linearly: maxseg per window 2017 * (maxseg^2 / cwnd per packet). 2018 */ 2019 if ((!tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) || 2020 !IN_FASTRECOVERY(tp)) { 2021 u_int cw = tp->snd_cwnd; 2022 u_int incr = tp->t_maxseg; 2023 if (cw > tp->snd_ssthresh) 2024 incr = incr * incr / cw; 2025 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2026 } 2027 SOCKBUF_LOCK(&so->so_snd); 2028 if (acked > so->so_snd.sb_cc) { 2029 tp->snd_wnd -= so->so_snd.sb_cc; 2030 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2031 ourfinisacked = 1; 2032 } else { 2033 sbdrop_locked(&so->so_snd, acked); 2034 tp->snd_wnd -= acked; 2035 ourfinisacked = 0; 2036 } 2037 sowwakeup_locked(so); 2038 /* Detect una wraparound. */ 2039 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2040 !IN_FASTRECOVERY(tp) && 2041 SEQ_GT(tp->snd_una, tp->snd_recover) && 2042 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2043 tp->snd_recover = th->th_ack - 1; 2044 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2045 IN_FASTRECOVERY(tp) && 2046 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2047 EXIT_FASTRECOVERY(tp); 2048 tp->snd_una = th->th_ack; 2049 if (tp->t_flags & TF_SACK_PERMIT) { 2050 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2051 tp->snd_recover = tp->snd_una; 2052 } 2053 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2054 tp->snd_nxt = tp->snd_una; 2055 2056 switch (tp->t_state) { 2057 2058 /* 2059 * In FIN_WAIT_1 STATE in addition to the processing 2060 * for the ESTABLISHED state if our FIN is now acknowledged 2061 * then enter FIN_WAIT_2. 2062 */ 2063 case TCPS_FIN_WAIT_1: 2064 if (ourfinisacked) { 2065 /* 2066 * If we can't receive any more 2067 * data, then closing user can proceed. 2068 * Starting the timer is contrary to the 2069 * specification, but if we don't get a FIN 2070 * we'll hang forever. 2071 * 2072 * XXXjl: 2073 * we should release the tp also, and use a 2074 * compressed state. 2075 */ 2076 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2077 int timeout; 2078 2079 soisdisconnected(so); 2080 timeout = (tcp_fast_finwait2_recycle) ? 2081 tcp_finwait2_timeout : tcp_maxidle; 2082 tcp_timer_activate(tp, TT_2MSL, timeout); 2083 } 2084 tp->t_state = TCPS_FIN_WAIT_2; 2085 } 2086 break; 2087 2088 /* 2089 * In CLOSING STATE in addition to the processing for 2090 * the ESTABLISHED state if the ACK acknowledges our FIN 2091 * then enter the TIME-WAIT state, otherwise ignore 2092 * the segment. 2093 */ 2094 case TCPS_CLOSING: 2095 if (ourfinisacked) { 2096 KASSERT(headlocked, ("%s: process_ACK: " 2097 "head not locked", __func__)); 2098 tcp_twstart(tp); 2099 INP_INFO_WUNLOCK(&tcbinfo); 2100 headlocked = 0; 2101 m_freem(m); 2102 return; 2103 } 2104 break; 2105 2106 /* 2107 * In LAST_ACK, we may still be waiting for data to drain 2108 * and/or to be acked, as well as for the ack of our FIN. 2109 * If our FIN is now acknowledged, delete the TCB, 2110 * enter the closed state and return. 2111 */ 2112 case TCPS_LAST_ACK: 2113 if (ourfinisacked) { 2114 KASSERT(headlocked, ("%s: process_ACK: " 2115 "tcp_close: head not locked", __func__)); 2116 tp = tcp_close(tp); 2117 goto drop; 2118 } 2119 break; 2120 } 2121 } 2122 2123 step6: 2124 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2125 INP_LOCK_ASSERT(tp->t_inpcb); 2126 2127 /* 2128 * Update window information. 2129 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2130 */ 2131 if ((thflags & TH_ACK) && 2132 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2133 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2134 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2135 /* keep track of pure window updates */ 2136 if (tlen == 0 && 2137 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2138 tcpstat.tcps_rcvwinupd++; 2139 tp->snd_wnd = tiwin; 2140 tp->snd_wl1 = th->th_seq; 2141 tp->snd_wl2 = th->th_ack; 2142 if (tp->snd_wnd > tp->max_sndwnd) 2143 tp->max_sndwnd = tp->snd_wnd; 2144 needoutput = 1; 2145 } 2146 2147 /* 2148 * Process segments with URG. 2149 */ 2150 if ((thflags & TH_URG) && th->th_urp && 2151 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2152 /* 2153 * This is a kludge, but if we receive and accept 2154 * random urgent pointers, we'll crash in 2155 * soreceive. It's hard to imagine someone 2156 * actually wanting to send this much urgent data. 2157 */ 2158 SOCKBUF_LOCK(&so->so_rcv); 2159 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2160 th->th_urp = 0; /* XXX */ 2161 thflags &= ~TH_URG; /* XXX */ 2162 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2163 goto dodata; /* XXX */ 2164 } 2165 /* 2166 * If this segment advances the known urgent pointer, 2167 * then mark the data stream. This should not happen 2168 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2169 * a FIN has been received from the remote side. 2170 * In these states we ignore the URG. 2171 * 2172 * According to RFC961 (Assigned Protocols), 2173 * the urgent pointer points to the last octet 2174 * of urgent data. We continue, however, 2175 * to consider it to indicate the first octet 2176 * of data past the urgent section as the original 2177 * spec states (in one of two places). 2178 */ 2179 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2180 tp->rcv_up = th->th_seq + th->th_urp; 2181 so->so_oobmark = so->so_rcv.sb_cc + 2182 (tp->rcv_up - tp->rcv_nxt) - 1; 2183 if (so->so_oobmark == 0) 2184 so->so_rcv.sb_state |= SBS_RCVATMARK; 2185 sohasoutofband(so); 2186 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2187 } 2188 SOCKBUF_UNLOCK(&so->so_rcv); 2189 /* 2190 * Remove out of band data so doesn't get presented to user. 2191 * This can happen independent of advancing the URG pointer, 2192 * but if two URG's are pending at once, some out-of-band 2193 * data may creep in... ick. 2194 */ 2195 if (th->th_urp <= (u_long)tlen && 2196 !(so->so_options & SO_OOBINLINE)) { 2197 /* hdr drop is delayed */ 2198 tcp_pulloutofband(so, th, m, drop_hdrlen); 2199 } 2200 } else { 2201 /* 2202 * If no out of band data is expected, 2203 * pull receive urgent pointer along 2204 * with the receive window. 2205 */ 2206 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2207 tp->rcv_up = tp->rcv_nxt; 2208 } 2209 dodata: /* XXX */ 2210 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2211 INP_LOCK_ASSERT(tp->t_inpcb); 2212 2213 /* 2214 * Process the segment text, merging it into the TCP sequencing queue, 2215 * and arranging for acknowledgment of receipt if necessary. 2216 * This process logically involves adjusting tp->rcv_wnd as data 2217 * is presented to the user (this happens in tcp_usrreq.c, 2218 * case PRU_RCVD). If a FIN has already been received on this 2219 * connection then we just ignore the text. 2220 */ 2221 if ((tlen || (thflags & TH_FIN)) && 2222 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2223 tcp_seq save_start = th->th_seq; 2224 m_adj(m, drop_hdrlen); /* delayed header drop */ 2225 /* 2226 * Insert segment which includes th into TCP reassembly queue 2227 * with control block tp. Set thflags to whether reassembly now 2228 * includes a segment with FIN. This handles the common case 2229 * inline (segment is the next to be received on an established 2230 * connection, and the queue is empty), avoiding linkage into 2231 * and removal from the queue and repetition of various 2232 * conversions. 2233 * Set DELACK for segments received in order, but ack 2234 * immediately when segments are out of order (so 2235 * fast retransmit can work). 2236 */ 2237 if (th->th_seq == tp->rcv_nxt && 2238 LIST_EMPTY(&tp->t_segq) && 2239 TCPS_HAVEESTABLISHED(tp->t_state)) { 2240 if (DELAY_ACK(tp)) 2241 tp->t_flags |= TF_DELACK; 2242 else 2243 tp->t_flags |= TF_ACKNOW; 2244 tp->rcv_nxt += tlen; 2245 thflags = th->th_flags & TH_FIN; 2246 tcpstat.tcps_rcvpack++; 2247 tcpstat.tcps_rcvbyte += tlen; 2248 ND6_HINT(tp); 2249 SOCKBUF_LOCK(&so->so_rcv); 2250 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2251 m_freem(m); 2252 else 2253 sbappendstream_locked(&so->so_rcv, m); 2254 /* NB: sorwakeup_locked() does an implicit unlock. */ 2255 sorwakeup_locked(so); 2256 } else { 2257 /* 2258 * XXX: Due to the header drop above "th" is 2259 * theoretically invalid by now. Fortunately 2260 * m_adj() doesn't actually frees any mbufs 2261 * when trimming from the head. 2262 */ 2263 thflags = tcp_reass(tp, th, &tlen, m); 2264 tp->t_flags |= TF_ACKNOW; 2265 } 2266 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2267 tcp_update_sack_list(tp, save_start, save_start + tlen); 2268 #if 0 2269 /* 2270 * Note the amount of data that peer has sent into 2271 * our window, in order to estimate the sender's 2272 * buffer size. 2273 * XXX: Unused. 2274 */ 2275 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2276 #endif 2277 } else { 2278 m_freem(m); 2279 thflags &= ~TH_FIN; 2280 } 2281 2282 /* 2283 * If FIN is received ACK the FIN and let the user know 2284 * that the connection is closing. 2285 */ 2286 if (thflags & TH_FIN) { 2287 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2288 socantrcvmore(so); 2289 /* 2290 * If connection is half-synchronized 2291 * (ie NEEDSYN flag on) then delay ACK, 2292 * so it may be piggybacked when SYN is sent. 2293 * Otherwise, since we received a FIN then no 2294 * more input can be expected, send ACK now. 2295 */ 2296 if (tp->t_flags & TF_NEEDSYN) 2297 tp->t_flags |= TF_DELACK; 2298 else 2299 tp->t_flags |= TF_ACKNOW; 2300 tp->rcv_nxt++; 2301 } 2302 switch (tp->t_state) { 2303 2304 /* 2305 * In SYN_RECEIVED and ESTABLISHED STATES 2306 * enter the CLOSE_WAIT state. 2307 */ 2308 case TCPS_SYN_RECEIVED: 2309 tp->t_starttime = ticks; 2310 /*FALLTHROUGH*/ 2311 case TCPS_ESTABLISHED: 2312 tp->t_state = TCPS_CLOSE_WAIT; 2313 break; 2314 2315 /* 2316 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2317 * enter the CLOSING state. 2318 */ 2319 case TCPS_FIN_WAIT_1: 2320 tp->t_state = TCPS_CLOSING; 2321 break; 2322 2323 /* 2324 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2325 * starting the time-wait timer, turning off the other 2326 * standard timers. 2327 */ 2328 case TCPS_FIN_WAIT_2: 2329 KASSERT(headlocked == 1, ("%s: dodata: " 2330 "TCP_FIN_WAIT_2: head not locked", __func__)); 2331 tcp_twstart(tp); 2332 INP_INFO_WUNLOCK(&tcbinfo); 2333 return; 2334 } 2335 } 2336 INP_INFO_WUNLOCK(&tcbinfo); 2337 headlocked = 0; 2338 #ifdef TCPDEBUG 2339 if (so->so_options & SO_DEBUG) 2340 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2341 &tcp_savetcp, 0); 2342 #endif 2343 2344 /* 2345 * Return any desired output. 2346 */ 2347 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2348 (void) tcp_output(tp); 2349 2350 check_delack: 2351 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2352 __func__)); 2353 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 2354 INP_LOCK_ASSERT(tp->t_inpcb); 2355 if (tp->t_flags & TF_DELACK) { 2356 tp->t_flags &= ~TF_DELACK; 2357 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2358 } 2359 INP_UNLOCK(tp->t_inpcb); 2360 return; 2361 2362 dropafterack: 2363 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2364 /* 2365 * Generate an ACK dropping incoming segment if it occupies 2366 * sequence space, where the ACK reflects our state. 2367 * 2368 * We can now skip the test for the RST flag since all 2369 * paths to this code happen after packets containing 2370 * RST have been dropped. 2371 * 2372 * In the SYN-RECEIVED state, don't send an ACK unless the 2373 * segment we received passes the SYN-RECEIVED ACK test. 2374 * If it fails send a RST. This breaks the loop in the 2375 * "LAND" DoS attack, and also prevents an ACK storm 2376 * between two listening ports that have been sent forged 2377 * SYN segments, each with the source address of the other. 2378 */ 2379 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2380 (SEQ_GT(tp->snd_una, th->th_ack) || 2381 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2382 rstreason = BANDLIM_RST_OPENPORT; 2383 goto dropwithreset; 2384 } 2385 #ifdef TCPDEBUG 2386 if (so->so_options & SO_DEBUG) 2387 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2388 &tcp_savetcp, 0); 2389 #endif 2390 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2391 INP_INFO_WUNLOCK(&tcbinfo); 2392 tp->t_flags |= TF_ACKNOW; 2393 (void) tcp_output(tp); 2394 INP_UNLOCK(tp->t_inpcb); 2395 m_freem(m); 2396 return; 2397 2398 dropwithreset: 2399 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2400 2401 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2402 2403 if (tp != NULL) 2404 INP_UNLOCK(tp->t_inpcb); 2405 if (headlocked) 2406 INP_INFO_WUNLOCK(&tcbinfo); 2407 return; 2408 2409 drop: 2410 /* 2411 * Drop space held by incoming segment and return. 2412 */ 2413 #ifdef TCPDEBUG 2414 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2415 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2416 &tcp_savetcp, 0); 2417 #endif 2418 if (tp != NULL) 2419 INP_UNLOCK(tp->t_inpcb); 2420 if (headlocked) 2421 INP_INFO_WUNLOCK(&tcbinfo); 2422 m_freem(m); 2423 return; 2424 } 2425 2426 /* 2427 * Issue RST and make ACK acceptable to originator of segment. 2428 * The mbuf must still include the original packet header. 2429 * tp may be NULL. 2430 */ 2431 static void 2432 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2433 int tlen, int rstreason) 2434 { 2435 struct ip *ip; 2436 #ifdef INET6 2437 struct ip6_hdr *ip6; 2438 #endif 2439 /* Don't bother if destination was broadcast/multicast. */ 2440 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2441 goto drop; 2442 #ifdef INET6 2443 if (mtod(m, struct ip *)->ip_v == 6) { 2444 ip6 = mtod(m, struct ip6_hdr *); 2445 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2446 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2447 goto drop; 2448 /* IPv6 anycast check is done at tcp6_input() */ 2449 } else 2450 #endif 2451 { 2452 ip = mtod(m, struct ip *); 2453 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2454 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2455 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2456 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2457 goto drop; 2458 } 2459 2460 /* Perform bandwidth limiting. */ 2461 if (badport_bandlim(rstreason) < 0) 2462 goto drop; 2463 2464 /* tcp_respond consumes the mbuf chain. */ 2465 if (th->th_flags & TH_ACK) { 2466 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2467 th->th_ack, TH_RST); 2468 } else { 2469 if (th->th_flags & TH_SYN) 2470 tlen++; 2471 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2472 (tcp_seq)0, TH_RST|TH_ACK); 2473 } 2474 return; 2475 drop: 2476 m_freem(m); 2477 return; 2478 } 2479 2480 /* 2481 * Parse TCP options and place in tcpopt. 2482 */ 2483 static void 2484 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2485 { 2486 int opt, optlen; 2487 2488 to->to_flags = 0; 2489 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2490 opt = cp[0]; 2491 if (opt == TCPOPT_EOL) 2492 break; 2493 if (opt == TCPOPT_NOP) 2494 optlen = 1; 2495 else { 2496 if (cnt < 2) 2497 break; 2498 optlen = cp[1]; 2499 if (optlen < 2 || optlen > cnt) 2500 break; 2501 } 2502 switch (opt) { 2503 case TCPOPT_MAXSEG: 2504 if (optlen != TCPOLEN_MAXSEG) 2505 continue; 2506 if (!(flags & TO_SYN)) 2507 continue; 2508 to->to_flags |= TOF_MSS; 2509 bcopy((char *)cp + 2, 2510 (char *)&to->to_mss, sizeof(to->to_mss)); 2511 to->to_mss = ntohs(to->to_mss); 2512 break; 2513 case TCPOPT_WINDOW: 2514 if (optlen != TCPOLEN_WINDOW) 2515 continue; 2516 if (!(flags & TO_SYN)) 2517 continue; 2518 to->to_flags |= TOF_SCALE; 2519 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2520 break; 2521 case TCPOPT_TIMESTAMP: 2522 if (optlen != TCPOLEN_TIMESTAMP) 2523 continue; 2524 to->to_flags |= TOF_TS; 2525 bcopy((char *)cp + 2, 2526 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2527 to->to_tsval = ntohl(to->to_tsval); 2528 bcopy((char *)cp + 6, 2529 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2530 to->to_tsecr = ntohl(to->to_tsecr); 2531 break; 2532 #ifdef TCP_SIGNATURE 2533 /* 2534 * XXX In order to reply to a host which has set the 2535 * TCP_SIGNATURE option in its initial SYN, we have to 2536 * record the fact that the option was observed here 2537 * for the syncache code to perform the correct response. 2538 */ 2539 case TCPOPT_SIGNATURE: 2540 if (optlen != TCPOLEN_SIGNATURE) 2541 continue; 2542 to->to_flags |= TOF_SIGNATURE; 2543 to->to_signature = cp + 2; 2544 break; 2545 #endif 2546 case TCPOPT_SACK_PERMITTED: 2547 if (optlen != TCPOLEN_SACK_PERMITTED) 2548 continue; 2549 if (!(flags & TO_SYN)) 2550 continue; 2551 if (!tcp_do_sack) 2552 continue; 2553 to->to_flags |= TOF_SACKPERM; 2554 break; 2555 case TCPOPT_SACK: 2556 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2557 continue; 2558 if (flags & TO_SYN) 2559 continue; 2560 to->to_flags |= TOF_SACK; 2561 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2562 to->to_sacks = cp + 2; 2563 tcpstat.tcps_sack_rcv_blocks++; 2564 break; 2565 default: 2566 continue; 2567 } 2568 } 2569 } 2570 2571 /* 2572 * Pull out of band byte out of a segment so 2573 * it doesn't appear in the user's data queue. 2574 * It is still reflected in the segment length for 2575 * sequencing purposes. 2576 */ 2577 static void 2578 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2579 int off) 2580 { 2581 int cnt = off + th->th_urp - 1; 2582 2583 while (cnt >= 0) { 2584 if (m->m_len > cnt) { 2585 char *cp = mtod(m, caddr_t) + cnt; 2586 struct tcpcb *tp = sototcpcb(so); 2587 2588 tp->t_iobc = *cp; 2589 tp->t_oobflags |= TCPOOB_HAVEDATA; 2590 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2591 m->m_len--; 2592 if (m->m_flags & M_PKTHDR) 2593 m->m_pkthdr.len--; 2594 return; 2595 } 2596 cnt -= m->m_len; 2597 m = m->m_next; 2598 if (m == NULL) 2599 break; 2600 } 2601 panic("tcp_pulloutofband"); 2602 } 2603 2604 /* 2605 * Collect new round-trip time estimate 2606 * and update averages and current timeout. 2607 */ 2608 static void 2609 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2610 { 2611 int delta; 2612 2613 INP_LOCK_ASSERT(tp->t_inpcb); 2614 2615 tcpstat.tcps_rttupdated++; 2616 tp->t_rttupdated++; 2617 if (tp->t_srtt != 0) { 2618 /* 2619 * srtt is stored as fixed point with 5 bits after the 2620 * binary point (i.e., scaled by 8). The following magic 2621 * is equivalent to the smoothing algorithm in rfc793 with 2622 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2623 * point). Adjust rtt to origin 0. 2624 */ 2625 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2626 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2627 2628 if ((tp->t_srtt += delta) <= 0) 2629 tp->t_srtt = 1; 2630 2631 /* 2632 * We accumulate a smoothed rtt variance (actually, a 2633 * smoothed mean difference), then set the retransmit 2634 * timer to smoothed rtt + 4 times the smoothed variance. 2635 * rttvar is stored as fixed point with 4 bits after the 2636 * binary point (scaled by 16). The following is 2637 * equivalent to rfc793 smoothing with an alpha of .75 2638 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2639 * rfc793's wired-in beta. 2640 */ 2641 if (delta < 0) 2642 delta = -delta; 2643 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2644 if ((tp->t_rttvar += delta) <= 0) 2645 tp->t_rttvar = 1; 2646 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2647 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2648 } else { 2649 /* 2650 * No rtt measurement yet - use the unsmoothed rtt. 2651 * Set the variance to half the rtt (so our first 2652 * retransmit happens at 3*rtt). 2653 */ 2654 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2655 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2656 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2657 } 2658 tp->t_rtttime = 0; 2659 tp->t_rxtshift = 0; 2660 2661 /* 2662 * the retransmit should happen at rtt + 4 * rttvar. 2663 * Because of the way we do the smoothing, srtt and rttvar 2664 * will each average +1/2 tick of bias. When we compute 2665 * the retransmit timer, we want 1/2 tick of rounding and 2666 * 1 extra tick because of +-1/2 tick uncertainty in the 2667 * firing of the timer. The bias will give us exactly the 2668 * 1.5 tick we need. But, because the bias is 2669 * statistical, we have to test that we don't drop below 2670 * the minimum feasible timer (which is 2 ticks). 2671 */ 2672 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2673 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2674 2675 /* 2676 * We received an ack for a packet that wasn't retransmitted; 2677 * it is probably safe to discard any error indications we've 2678 * received recently. This isn't quite right, but close enough 2679 * for now (a route might have failed after we sent a segment, 2680 * and the return path might not be symmetrical). 2681 */ 2682 tp->t_softerror = 0; 2683 } 2684 2685 /* 2686 * Determine a reasonable value for maxseg size. 2687 * If the route is known, check route for mtu. 2688 * If none, use an mss that can be handled on the outgoing 2689 * interface without forcing IP to fragment; if bigger than 2690 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2691 * to utilize large mbufs. If no route is found, route has no mtu, 2692 * or the destination isn't local, use a default, hopefully conservative 2693 * size (usually 512 or the default IP max size, but no more than the mtu 2694 * of the interface), as we can't discover anything about intervening 2695 * gateways or networks. We also initialize the congestion/slow start 2696 * window to be a single segment if the destination isn't local. 2697 * While looking at the routing entry, we also initialize other path-dependent 2698 * parameters from pre-set or cached values in the routing entry. 2699 * 2700 * Also take into account the space needed for options that we 2701 * send regularly. Make maxseg shorter by that amount to assure 2702 * that we can send maxseg amount of data even when the options 2703 * are present. Store the upper limit of the length of options plus 2704 * data in maxopd. 2705 * 2706 * In case of T/TCP, we call this routine during implicit connection 2707 * setup as well (offer = -1), to initialize maxseg from the cached 2708 * MSS of our peer. 2709 * 2710 * NOTE that this routine is only called when we process an incoming 2711 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2712 */ 2713 void 2714 tcp_mss(struct tcpcb *tp, int offer) 2715 { 2716 int rtt, mss; 2717 u_long bufsize; 2718 u_long maxmtu; 2719 struct inpcb *inp = tp->t_inpcb; 2720 struct socket *so; 2721 struct hc_metrics_lite metrics; 2722 int origoffer = offer; 2723 int mtuflags = 0; 2724 #ifdef INET6 2725 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2726 size_t min_protoh = isipv6 ? 2727 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2728 sizeof (struct tcpiphdr); 2729 #else 2730 const size_t min_protoh = sizeof(struct tcpiphdr); 2731 #endif 2732 2733 /* Initialize. */ 2734 #ifdef INET6 2735 if (isipv6) { 2736 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2737 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2738 } else 2739 #endif 2740 { 2741 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2742 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2743 } 2744 so = inp->inp_socket; 2745 2746 /* 2747 * No route to sender, stay with default mss and return. 2748 */ 2749 if (maxmtu == 0) 2750 return; 2751 2752 /* What have we got? */ 2753 switch (offer) { 2754 case 0: 2755 /* 2756 * Offer == 0 means that there was no MSS on the SYN 2757 * segment, in this case we use tcp_mssdflt. 2758 */ 2759 offer = 2760 #ifdef INET6 2761 isipv6 ? tcp_v6mssdflt : 2762 #endif 2763 tcp_mssdflt; 2764 break; 2765 2766 case -1: 2767 /* 2768 * Offer == -1 means that we didn't receive SYN yet. 2769 */ 2770 /* FALLTHROUGH */ 2771 2772 default: 2773 /* 2774 * Prevent DoS attack with too small MSS. Round up 2775 * to at least minmss. 2776 */ 2777 offer = max(offer, tcp_minmss); 2778 /* 2779 * Sanity check: make sure that maxopd will be large 2780 * enough to allow some data on segments even if the 2781 * all the option space is used (40bytes). Otherwise 2782 * funny things may happen in tcp_output. 2783 */ 2784 offer = max(offer, 64); 2785 } 2786 2787 /* 2788 * rmx information is now retrieved from tcp_hostcache. 2789 */ 2790 tcp_hc_get(&inp->inp_inc, &metrics); 2791 2792 /* 2793 * If there's a discovered mtu int tcp hostcache, use it 2794 * else, use the link mtu. 2795 */ 2796 if (metrics.rmx_mtu) 2797 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2798 else { 2799 #ifdef INET6 2800 if (isipv6) { 2801 mss = maxmtu - min_protoh; 2802 if (!path_mtu_discovery && 2803 !in6_localaddr(&inp->in6p_faddr)) 2804 mss = min(mss, tcp_v6mssdflt); 2805 } else 2806 #endif 2807 { 2808 mss = maxmtu - min_protoh; 2809 if (!path_mtu_discovery && 2810 !in_localaddr(inp->inp_faddr)) 2811 mss = min(mss, tcp_mssdflt); 2812 } 2813 } 2814 mss = min(mss, offer); 2815 2816 /* 2817 * maxopd stores the maximum length of data AND options 2818 * in a segment; maxseg is the amount of data in a normal 2819 * segment. We need to store this value (maxopd) apart 2820 * from maxseg, because now every segment carries options 2821 * and thus we normally have somewhat less data in segments. 2822 */ 2823 tp->t_maxopd = mss; 2824 2825 /* 2826 * origoffer==-1 indicates that no segments were received yet. 2827 * In this case we just guess. 2828 */ 2829 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2830 (origoffer == -1 || 2831 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2832 mss -= TCPOLEN_TSTAMP_APPA; 2833 tp->t_maxseg = mss; 2834 2835 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2836 if (mss > MCLBYTES) 2837 mss &= ~(MCLBYTES-1); 2838 #else 2839 if (mss > MCLBYTES) 2840 mss = mss / MCLBYTES * MCLBYTES; 2841 #endif 2842 tp->t_maxseg = mss; 2843 2844 /* 2845 * If there's a pipesize, change the socket buffer to that size, 2846 * don't change if sb_hiwat is different than default (then it 2847 * has been changed on purpose with setsockopt). 2848 * Make the socket buffers an integral number of mss units; 2849 * if the mss is larger than the socket buffer, decrease the mss. 2850 */ 2851 SOCKBUF_LOCK(&so->so_snd); 2852 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2853 bufsize = metrics.rmx_sendpipe; 2854 else 2855 bufsize = so->so_snd.sb_hiwat; 2856 if (bufsize < mss) 2857 mss = bufsize; 2858 else { 2859 bufsize = roundup(bufsize, mss); 2860 if (bufsize > sb_max) 2861 bufsize = sb_max; 2862 if (bufsize > so->so_snd.sb_hiwat) 2863 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2864 } 2865 SOCKBUF_UNLOCK(&so->so_snd); 2866 tp->t_maxseg = mss; 2867 2868 SOCKBUF_LOCK(&so->so_rcv); 2869 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2870 bufsize = metrics.rmx_recvpipe; 2871 else 2872 bufsize = so->so_rcv.sb_hiwat; 2873 if (bufsize > mss) { 2874 bufsize = roundup(bufsize, mss); 2875 if (bufsize > sb_max) 2876 bufsize = sb_max; 2877 if (bufsize > so->so_rcv.sb_hiwat) 2878 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2879 } 2880 SOCKBUF_UNLOCK(&so->so_rcv); 2881 /* 2882 * While we're here, check the others too. 2883 */ 2884 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2885 tp->t_srtt = rtt; 2886 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2887 tcpstat.tcps_usedrtt++; 2888 if (metrics.rmx_rttvar) { 2889 tp->t_rttvar = metrics.rmx_rttvar; 2890 tcpstat.tcps_usedrttvar++; 2891 } else { 2892 /* default variation is +- 1 rtt */ 2893 tp->t_rttvar = 2894 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2895 } 2896 TCPT_RANGESET(tp->t_rxtcur, 2897 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2898 tp->t_rttmin, TCPTV_REXMTMAX); 2899 } 2900 if (metrics.rmx_ssthresh) { 2901 /* 2902 * There's some sort of gateway or interface 2903 * buffer limit on the path. Use this to set 2904 * the slow start threshhold, but set the 2905 * threshold to no less than 2*mss. 2906 */ 2907 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2908 tcpstat.tcps_usedssthresh++; 2909 } 2910 if (metrics.rmx_bandwidth) 2911 tp->snd_bandwidth = metrics.rmx_bandwidth; 2912 2913 /* 2914 * Set the slow-start flight size depending on whether this 2915 * is a local network or not. 2916 * 2917 * Extend this so we cache the cwnd too and retrieve it here. 2918 * Make cwnd even bigger than RFC3390 suggests but only if we 2919 * have previous experience with the remote host. Be careful 2920 * not make cwnd bigger than remote receive window or our own 2921 * send socket buffer. Maybe put some additional upper bound 2922 * on the retrieved cwnd. Should do incremental updates to 2923 * hostcache when cwnd collapses so next connection doesn't 2924 * overloads the path again. 2925 * 2926 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2927 * We currently check only in syncache_socket for that. 2928 */ 2929 #define TCP_METRICS_CWND 2930 #ifdef TCP_METRICS_CWND 2931 if (metrics.rmx_cwnd) 2932 tp->snd_cwnd = max(mss, 2933 min(metrics.rmx_cwnd / 2, 2934 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2935 else 2936 #endif 2937 if (tcp_do_rfc3390) 2938 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2939 #ifdef INET6 2940 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2941 (!isipv6 && in_localaddr(inp->inp_faddr))) 2942 #else 2943 else if (in_localaddr(inp->inp_faddr)) 2944 #endif 2945 tp->snd_cwnd = mss * ss_fltsz_local; 2946 else 2947 tp->snd_cwnd = mss * ss_fltsz; 2948 2949 /* Check the interface for TSO capabilities. */ 2950 if (mtuflags & CSUM_TSO) 2951 tp->t_flags |= TF_TSO; 2952 } 2953 2954 /* 2955 * Determine the MSS option to send on an outgoing SYN. 2956 */ 2957 int 2958 tcp_mssopt(struct in_conninfo *inc) 2959 { 2960 int mss = 0; 2961 u_long maxmtu = 0; 2962 u_long thcmtu = 0; 2963 size_t min_protoh; 2964 #ifdef INET6 2965 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2966 #endif 2967 2968 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2969 2970 #ifdef INET6 2971 if (isipv6) { 2972 mss = tcp_v6mssdflt; 2973 maxmtu = tcp_maxmtu6(inc, NULL); 2974 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2975 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2976 } else 2977 #endif 2978 { 2979 mss = tcp_mssdflt; 2980 maxmtu = tcp_maxmtu(inc, NULL); 2981 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2982 min_protoh = sizeof(struct tcpiphdr); 2983 } 2984 if (maxmtu && thcmtu) 2985 mss = min(maxmtu, thcmtu) - min_protoh; 2986 else if (maxmtu || thcmtu) 2987 mss = max(maxmtu, thcmtu) - min_protoh; 2988 2989 return (mss); 2990 } 2991 2992 2993 /* 2994 * On a partial ack arrives, force the retransmission of the 2995 * next unacknowledged segment. Do not clear tp->t_dupacks. 2996 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2997 * be started again. 2998 */ 2999 static void 3000 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3001 { 3002 tcp_seq onxt = tp->snd_nxt; 3003 u_long ocwnd = tp->snd_cwnd; 3004 3005 tcp_timer_activate(tp, TT_REXMT, 0); 3006 tp->t_rtttime = 0; 3007 tp->snd_nxt = th->th_ack; 3008 /* 3009 * Set snd_cwnd to one segment beyond acknowledged offset. 3010 * (tp->snd_una has not yet been updated when this function is called.) 3011 */ 3012 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3013 tp->t_flags |= TF_ACKNOW; 3014 (void) tcp_output(tp); 3015 tp->snd_cwnd = ocwnd; 3016 if (SEQ_GT(onxt, tp->snd_nxt)) 3017 tp->snd_nxt = onxt; 3018 /* 3019 * Partial window deflation. Relies on fact that tp->snd_una 3020 * not updated yet. 3021 */ 3022 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3023 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3024 else 3025 tp->snd_cwnd = 0; 3026 tp->snd_cwnd += tp->t_maxseg; 3027 } 3028